---

📦 Plugins Overview

Plugin	Feature	Description
📄 dsh-file-preview	In-chat File Preview	Right-click file paths in chat to preview contents directly in the side panel — no more jumping to external apps
🖼️ dsh-image-analyzer	Dual-Model Vision	Pair a text-only main model (e.g. DeepSeek) with a vision-capable LLM so images can be understood

---

✨ Features

📄 dsh-file-preview

• ✅ Right-click any file path reference in chat → choose "Preview in right panel"
• ✅ Supports: text files (syntax highlighted) / images / PDF
• ✅ Keeps the original tool-details panel intact
• ✅ Resizable preview panel

🖼️ dsh-image-analyzer

• ✅ When the main model can't see images, it calls the analyze_image tool
• ✅ The tool sends the image to a vision-capable model (default agnes-2.5-flash, OpenAI-compatible)
• ✅ Returns a text description so the main model can continue the task
• ✅ Works with any OpenAI-compatible vision model: GPT-4o / Qwen-VL / GLM-4V ...

---

🚀 Quick Install

Prerequisites

• DeepSeek Harness Web (dsh web)
• Locate your web profile directory (default: <install-dir>/data/profiles/web)

One-Click Install

Windows (admin PowerShell):

.\install.ps1

Linux / macOS:

chmod +x install.sh && ./install.sh

The script auto-detects the profile directory, copies plugins, creates links, and updates config.

Manual Install (3 steps)

① Copy plugins & create links:

cp -r dsh-file-preview dsh-image-analyzer <PROFILE_DIR>/
ln -s <PROFILE_DIR>/dsh-file-preview <PROFILE_DIR>/../node_modules/dsh-file-preview
ln -s <PROFILE_DIR>/dsh-image-analyzer <PROFILE_DIR>/../node_modules/dsh-image-analyzer

② Register plugins (<PROFILE_DIR>/cordis.patch.yml):

- insert:
    - id: dsh-file-preview
      name: 'dsh-file-preview'
    - id: dsh-image-analyzer
      name: 'dsh-image-analyzer'

③ Register the vision tool (~/.dsh/.agent-presets/<your-preset>/agent.cordis.yml):

- id: image-analyzer
  name: 'dsh-image-analyzer'

④ Configure the vision model API key (~/.dsh/settings.yaml, see example)

⑤ Restart DSH and create a new session 🎉

---

⚙️ Configuration

dsh-image-analyzer vision model config

Option	Default	Description
baseURL	https://apihub.agnes-ai.com/v1	Vision model API endpoint (OpenAI-compatible)
model	agnes-2.5-flash	Vision model ID
apiKeyEnv	AGNES_API_KEY	Env/credential name of the API key
maxBytes	10485760	Max image bytes (10 MB)

Supported formats: PNG / JPG / JPEG / WebP / GIF

---

🛠 How It Works

【File Preview】
browser client.js ──right-click detect──▶ fetch('/api/file.read') ──▶ host index.js ──▶ file content
【Vision Analysis】
main model (DeepSeek) ──calls analyze_image──▶ plugin reads image ──base64──▶ vision model API ──▶ text description

---

📜 License

MIT — free to use, modify, and share